Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77064.71884409753,
-      "required_pages_per_second_upper_estimate": 0.005331516064418255
+      "required_token_positions_per_second": 76785.34468962895,
+      "required_pages_per_second_upper_estimate": 0.005312188312174771
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9633.089855512191,
-      "required_pages_per_second_upper_estimate": 0.0040331735607119015
+      "required_token_positions_per_second": 9598.16808620362,
+      "required_pages_per_second_upper_estimate": 0.0040185525451518714
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19266.179711024382,
-      "required_pages_per_second_upper_estimate": 0.004012637150480275
+      "required_token_positions_per_second": 19196.33617240724,
+      "required_pages_per_second_upper_estimate": 0.0039980905832842995
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28899.26956653657,
-      "required_pages_per_second_upper_estimate": 0.001732467083884254
+      "required_token_positions_per_second": 28794.504258610854,
+      "required_pages_per_second_upper_estimate": 0.00172618656364147
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28899.26956653657,
-      "required_pages_per_second_upper_estimate": 0.002246791094345265
+      "required_token_positions_per_second": 28794.504258610854,
+      "required_pages_per_second_upper_estimate": 0.0022386460524679293
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28899.26956653657,
-      "required_pages_per_second_upper_estimate": 0.3706888003814288
+      "required_token_positions_per_second": 28794.504258610854,
+      "required_pages_per_second_upper_estimate": 0.3693449834995812
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.17206636879689086,
+    "producer_batch_seconds": 0.10534007390637612,
     "producer_p95_seconds": 0.22015652299887734,
-    "gpu_update_seconds": 0.9567023808800531,
-    "producer_batches_per_second": 5.8117109519548915,
-    "gpu_batches_per_second": 1.0452571457804027,
-    "producer_headroom": 5.560077704721925,
+    "gpu_update_seconds": 0.9601832263436868,
+    "producer_batches_per_second": 9.493063398538883,
+    "gpu_batches_per_second": 1.0414678912981357,
+    "producer_headroom": 9.11508024189423,
     "sustainable": true,
-    "buffer_seconds": 60.272249995443346,
+    "buffer_seconds": 60.49154325965227,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 534808.5683044991,
+    "generation_bytes_per_second": 556179.130581579,
     "upload_bytes_per_second": 5168575.3491557585,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.75746905184965,
+    "estimated_full_shards_per_hour": 119.34309423528222,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -162,3 +162,4 @@
 {"step": 620600, "loss": 0.5208194851875305, "lr": 0.0001, "grad_norm": 0.14689715045845486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90839094921, "tokens_per_second": 151449.25736804775, "data_wait_s": 1.42680000863038e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 620700, "loss": 0.5945292115211487, "lr": 0.0001, "grad_norm": 0.17012109593468946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90853791441, "tokens_per_second": 152786.06384667708, "data_wait_s": 1.4075998478801921e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 620800, "loss": 0.5377899408340454, "lr": 0.0001, "grad_norm": 0.17761723371324462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90868425634, "tokens_per_second": 152212.34361118838, "data_wait_s": 1.4498000382445753e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 620900, "loss": 0.5788757801055908, "lr": 0.0001, "grad_norm": 0.1803991948934275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90883023139, "tokens_per_second": 151778.5397582285, "data_wait_s": 9.416999091627076e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a34eaa23ca630b0226aa9cdd18b5b41ef280870ac136a2eeb7c27cd44982592a
-size 350649
+oid sha256:6af4b4293323f2bc0e3d2f1fe7bf269da46ea68c1763a0e825a86fbd7d3166a3
+size 352767